1 // SPDX-License-Identifier: GPL-2.0
6 * YOSHIFUJI Hideaki @USAGI
7 * Split up af-specific portion
11 #include <linux/bottom_half.h>
12 #include <linux/cache.h>
13 #include <linux/interrupt.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/netdevice.h>
17 #include <linux/percpu.h>
21 #include <net/ip_tunnels.h>
22 #include <net/ip6_tunnel.h>
23 #include <net/dst_metadata.h>
25 #include "xfrm_inout.h"
27 struct xfrm_trans_tasklet {
28 struct work_struct work;
29 spinlock_t queue_lock;
30 struct sk_buff_head queue;
33 struct xfrm_trans_cb {
35 struct inet_skb_parm h4;
36 #if IS_ENABLED(CONFIG_IPV6)
37 struct inet6_skb_parm h6;
40 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
44 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
46 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
47 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
49 static struct gro_cells gro_cells;
50 static struct net_device xfrm_napi_dev;
52 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
54 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
58 if (WARN_ON(afinfo->family > AF_INET6))
61 spin_lock_bh(&xfrm_input_afinfo_lock);
62 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
65 rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
66 spin_unlock_bh(&xfrm_input_afinfo_lock);
69 EXPORT_SYMBOL(xfrm_input_register_afinfo);
71 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
75 spin_lock_bh(&xfrm_input_afinfo_lock);
76 if (likely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family])) {
77 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family] != afinfo))
80 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
82 spin_unlock_bh(&xfrm_input_afinfo_lock);
86 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
88 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
90 const struct xfrm_input_afinfo *afinfo;
92 if (WARN_ON_ONCE(family > AF_INET6))
96 afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
97 if (unlikely(!afinfo))
102 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
105 bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
106 const struct xfrm_input_afinfo *afinfo;
109 afinfo = xfrm_input_get_afinfo(family, is_ipip);
111 return -EAFNOSUPPORT;
113 ret = afinfo->callback(skb, protocol, err);
119 struct sec_path *secpath_set(struct sk_buff *skb)
121 struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
123 sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
127 if (tmp) /* reused existing one (was COW'd if needed) */
130 /* allocated new secpath */
131 memset(sp->ovec, 0, sizeof(sp->ovec));
134 sp->verified_cnt = 0;
138 EXPORT_SYMBOL(secpath_set);
140 /* Fetch spi and seq from ipsec header */
142 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
144 int offset, offset_seq;
149 hlen = sizeof(struct ip_auth_hdr);
150 offset = offsetof(struct ip_auth_hdr, spi);
151 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
154 hlen = sizeof(struct ip_esp_hdr);
155 offset = offsetof(struct ip_esp_hdr, spi);
156 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
159 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
161 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
168 if (!pskb_may_pull(skb, hlen))
171 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
172 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
175 EXPORT_SYMBOL(xfrm_parse_spi);
177 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
183 skb->protocol = htons(ETH_P_IP);
185 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
186 struct ip_beet_phdr *ph;
189 if (!pskb_may_pull(skb, sizeof(*ph)))
192 ph = (struct ip_beet_phdr *)skb->data;
194 phlen = sizeof(*ph) + ph->padlen;
195 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
196 if (optlen < 0 || optlen & 3 || optlen > 250)
199 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
201 if (!pskb_may_pull(skb, phlen))
203 __skb_pull(skb, phlen);
206 skb_push(skb, sizeof(*iph));
207 skb_reset_network_header(skb);
208 skb_mac_header_rebuild(skb);
210 xfrm4_beet_make_header(skb);
214 iph->ihl += optlen / 4;
215 iph->tot_len = htons(skb->len);
216 iph->daddr = x->sel.daddr.a4;
217 iph->saddr = x->sel.saddr.a4;
219 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
225 static void ipip_ecn_decapsulate(struct sk_buff *skb)
227 struct iphdr *inner_iph = ipip_hdr(skb);
229 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
230 IP_ECN_set_ce(inner_iph);
233 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
237 skb->protocol = htons(ETH_P_IP);
239 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
242 err = skb_unclone(skb, GFP_ATOMIC);
246 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
247 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
248 if (!(x->props.flags & XFRM_STATE_NOECN))
249 ipip_ecn_decapsulate(skb);
251 skb_reset_network_header(skb);
252 skb_mac_header_rebuild(skb);
254 eth_hdr(skb)->h_proto = skb->protocol;
262 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
264 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
266 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
267 IP6_ECN_set_ce(skb, inner_iph);
270 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
274 skb->protocol = htons(ETH_P_IPV6);
276 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
279 err = skb_unclone(skb, GFP_ATOMIC);
283 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
284 ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
285 if (!(x->props.flags & XFRM_STATE_NOECN))
286 ipip6_ecn_decapsulate(skb);
288 skb_reset_network_header(skb);
289 skb_mac_header_rebuild(skb);
291 eth_hdr(skb)->h_proto = skb->protocol;
299 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
301 struct ipv6hdr *ip6h;
302 int size = sizeof(struct ipv6hdr);
305 skb->protocol = htons(ETH_P_IPV6);
307 err = skb_cow_head(skb, size + skb->mac_len);
311 __skb_push(skb, size);
312 skb_reset_network_header(skb);
313 skb_mac_header_rebuild(skb);
315 xfrm6_beet_make_header(skb);
317 ip6h = ipv6_hdr(skb);
318 ip6h->payload_len = htons(skb->len - size);
319 ip6h->daddr = x->sel.daddr.in6;
320 ip6h->saddr = x->sel.saddr.in6;
326 /* Remove encapsulation header.
328 * The IP header will be moved over the top of the encapsulation
331 * On entry, the transport header shall point to where the IP header
332 * should be and the network header shall be set to where the IP
333 * header currently is. skb->data shall point to the start of the
337 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
340 switch (x->props.mode) {
342 switch (x->sel.family) {
344 return xfrm4_remove_beet_encap(x, skb);
346 return xfrm6_remove_beet_encap(x, skb);
349 case XFRM_MODE_TUNNEL:
350 switch (XFRM_MODE_SKB_CB(skb)->protocol) {
352 return xfrm4_remove_tunnel_encap(x, skb);
354 return xfrm6_remove_tunnel_encap(x, skb);
364 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
366 switch (x->props.family) {
368 xfrm4_extract_header(skb);
371 xfrm6_extract_header(skb);
375 return -EAFNOSUPPORT;
378 return xfrm_inner_mode_encap_remove(x, skb);
381 /* Remove encapsulation header.
383 * The IP header will be moved over the top of the encapsulation header.
385 * On entry, skb_transport_header() shall point to where the IP header
386 * should be and skb_network_header() shall be set to where the IP header
387 * currently is. skb->data shall point to the start of the payload.
389 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
391 struct xfrm_offload *xo = xfrm_offload(skb);
392 int ihl = skb->data - skb_transport_header(skb);
394 if (skb->transport_header != skb->network_header) {
395 memmove(skb_transport_header(skb),
396 skb_network_header(skb), ihl);
399 skb_mac_header_was_set(skb) ? skb_mac_header_len(skb) : 0;
400 skb->network_header = skb->transport_header;
402 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
403 skb_reset_transport_header(skb);
407 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
409 #if IS_ENABLED(CONFIG_IPV6)
410 struct xfrm_offload *xo = xfrm_offload(skb);
411 int ihl = skb->data - skb_transport_header(skb);
413 if (skb->transport_header != skb->network_header) {
414 memmove(skb_transport_header(skb),
415 skb_network_header(skb), ihl);
418 skb_mac_header_was_set(skb) ? skb_mac_header_len(skb) : 0;
419 skb->network_header = skb->transport_header;
421 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
422 sizeof(struct ipv6hdr));
423 skb_reset_transport_header(skb);
427 return -EAFNOSUPPORT;
431 static int xfrm_inner_mode_input(struct xfrm_state *x,
434 switch (x->props.mode) {
436 case XFRM_MODE_TUNNEL:
437 return xfrm_prepare_input(x, skb);
438 case XFRM_MODE_TRANSPORT:
439 if (x->props.family == AF_INET)
440 return xfrm4_transport_input(x, skb);
441 if (x->props.family == AF_INET6)
442 return xfrm6_transport_input(x, skb);
444 case XFRM_MODE_ROUTEOPTIMIZATION:
455 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
457 const struct xfrm_state_afinfo *afinfo;
458 struct net *net = dev_net(skb->dev);
462 struct xfrm_state *x = NULL;
463 xfrm_address_t *daddr;
464 u32 mark = skb->mark;
465 unsigned int family = AF_UNSPEC;
468 bool xfrm_gro = false;
469 bool crypto_done = false;
470 struct xfrm_offload *xo = xfrm_offload(skb);
473 if (encap_type < 0) {
474 x = xfrm_input_state(skb);
476 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
477 if (x->km.state == XFRM_STATE_ACQ)
478 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
481 LINUX_MIB_XFRMINSTATEINVALID);
483 if (encap_type == -1)
488 family = x->props.family;
490 /* An encap_type of -1 indicates async resumption. */
491 if (encap_type == -1) {
493 seq = XFRM_SKB_CB(skb)->seq.input.low;
497 /* encap_type < -1 indicates a GRO call. */
499 seq = XFRM_SPI_SKB_CB(skb)->seq;
501 if (xo && (xo->flags & CRYPTO_DONE)) {
503 family = XFRM_SPI_SKB_CB(skb)->family;
505 if (!(xo->status & CRYPTO_SUCCESS)) {
507 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
508 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
509 CRYPTO_TUNNEL_AH_AUTH_FAILED |
510 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
512 xfrm_audit_state_icvfail(x, skb,
514 x->stats.integrity_failed++;
515 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
519 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
520 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
524 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
528 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
529 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
537 family = XFRM_SPI_SKB_CB(skb)->family;
539 /* if tunnel is present override skb->mark value with tunnel i_key */
542 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
543 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
546 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
547 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
551 sp = secpath_set(skb);
553 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
558 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
560 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
564 daddr = (xfrm_address_t *)(skb_network_header(skb) +
565 XFRM_SPI_SKB_CB(skb)->daddroff);
567 sp = skb_sec_path(skb);
569 if (sp->len == XFRM_MAX_DEPTH) {
571 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
575 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
578 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
579 xfrm_audit_state_notfound(skb, family, spi, seq);
583 skb->mark = xfrm_smark_get(skb->mark, x);
585 sp->xvec[sp->len++] = x;
589 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
596 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
597 if (x->km.state == XFRM_STATE_ACQ)
598 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
601 LINUX_MIB_XFRMINSTATEINVALID);
605 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
606 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
610 if (xfrm_replay_check(x, skb, seq)) {
611 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
615 if (xfrm_state_check_expire(x)) {
616 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
620 spin_unlock(&x->lock);
622 if (xfrm_tunnel_check(skb, x, family)) {
623 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
627 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
629 XFRM_SKB_CB(skb)->seq.input.low = seq;
630 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
635 nexthdr = x->type_offload->input_tail(x, skb);
637 nexthdr = x->type->input(x, skb);
639 if (nexthdr == -EINPROGRESS)
646 if (nexthdr == -EBADMSG) {
647 xfrm_audit_state_icvfail(x, skb,
649 x->stats.integrity_failed++;
651 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
655 /* only the first xfrm gets the encap type */
658 if (xfrm_replay_recheck(x, skb, seq)) {
659 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
663 xfrm_replay_advance(x, seq);
665 x->curlft.bytes += skb->len;
668 spin_unlock(&x->lock);
670 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
672 if (xfrm_inner_mode_input(x, skb)) {
673 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
677 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
683 * We need the inner address. However, we only get here for
684 * transport mode so the outer address is identical.
686 daddr = &x->id.daddr;
687 family = x->props.family;
689 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
691 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
697 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
704 sp = skb_sec_path(skb);
707 if (skb_valid_dst(skb))
709 gro_cells_receive(&gro_cells, skb);
712 xo = xfrm_offload(skb);
714 xfrm_gro = xo->flags & XFRM_GRO;
718 afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
720 err = afinfo->transport_finish(skb, xfrm_gro || async);
723 sp = skb_sec_path(skb);
726 if (skb_valid_dst(skb))
728 gro_cells_receive(&gro_cells, skb);
736 spin_unlock(&x->lock);
738 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
742 EXPORT_SYMBOL(xfrm_input);
744 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
746 return xfrm_input(skb, nexthdr, 0, -1);
748 EXPORT_SYMBOL(xfrm_input_resume);
750 static void xfrm_trans_reinject(struct work_struct *work)
752 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
753 struct sk_buff_head queue;
756 __skb_queue_head_init(&queue);
757 spin_lock_bh(&trans->queue_lock);
758 skb_queue_splice_init(&trans->queue, &queue);
759 spin_unlock_bh(&trans->queue_lock);
762 while ((skb = __skb_dequeue(&queue)))
763 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
768 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
769 int (*finish)(struct net *, struct sock *,
772 struct xfrm_trans_tasklet *trans;
774 trans = this_cpu_ptr(&xfrm_trans_tasklet);
776 if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
779 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
781 XFRM_TRANS_SKB_CB(skb)->finish = finish;
782 XFRM_TRANS_SKB_CB(skb)->net = net;
783 spin_lock_bh(&trans->queue_lock);
784 __skb_queue_tail(&trans->queue, skb);
785 spin_unlock_bh(&trans->queue_lock);
786 schedule_work(&trans->work);
789 EXPORT_SYMBOL(xfrm_trans_queue_net);
791 int xfrm_trans_queue(struct sk_buff *skb,
792 int (*finish)(struct net *, struct sock *,
795 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
797 EXPORT_SYMBOL(xfrm_trans_queue);
799 void __init xfrm_input_init(void)
804 init_dummy_netdev(&xfrm_napi_dev);
805 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
807 gro_cells.cells = NULL;
809 for_each_possible_cpu(i) {
810 struct xfrm_trans_tasklet *trans;
812 trans = &per_cpu(xfrm_trans_tasklet, i);
813 spin_lock_init(&trans->queue_lock);
814 __skb_queue_head_init(&trans->queue);
815 INIT_WORK(&trans->work, xfrm_trans_reinject);